Flexible Cutoff Learning: Optimizing Machine Learning Potentials After Training

TL;DR

Flexible Cutoff Learning (FCL) allows machine learning potentials to have adjustable cutoff radii post-training, enabling tailored accuracy-cost tradeoffs for different applications without retraining.

Contribution

FCL introduces a novel training method where cutoff radii are sampled during training, allowing post-training optimization of per-atom cutoffs for diverse systems.

Findings

Reduces computational cost by over 60% in molecular crystal applications.

Maintains force errors below 1% despite cutoff optimization.

Enables a single MLIP to adapt to multiple applications without retraining.

Abstract

We introduce Flexible Cutoff Learning (FCL), a method for training machine learning interatomic potentials (MLIPs) whose cutoff radii can be adjusted after training. Unlike conventional MLIPs that fix the cutoff radius during training, FCL models are trained by randomly sampling cutoff radii independently for each atom. The resulting model can then be deployed with different per-atom cutoff radii depending on the application, enabling application-specific optimization of the accuracy-cost tradeoff. Using a differentiable cost model, these per-atom cutoffs can be optimized for specific target systems after training. We demonstrate FCL with a modified MACE architecture trained on the MAD dataset. For a subset featuring molecular crystals, optimized per-atom cutoffs reduce computational cost by more than 60% while increasing force errors by less than 1%. These results show that FCL enables training of a single general-purpose MLIP that can be adapted to diverse applications through post-training cutoff optimization, eliminating the need for retraining.